Goal of this project

Human-like locomotion is a complex control problem, as it requires both robustness and stylized motion. Reinforcement learning (RL) has been widely adopted to address model uncertainties and environmental discrepancies, achieving robust locomotion. However, the black-box nature of neural networks hampers the debugging process, and stylized motions often require tedious reward shaping. In this regard, motion priors can be an effective alternative to mitigate these challenges. Specifically, AMP (Adversarial Motion Priors) provides style rewards based on reference motion data. However, most motion datasets do not include toe joints, leading to unnecessary toe movements or ineffective toe usage during locomotion. Therefore, the goal of this project is to develop a method that enables relatively small but functionally important toe joints to be utilized naturally and effectively, even when they are absent from the motion prior data.